Materials Today Bio 23 (2023) 100792

Contents lists available at ScienceDirect

Materials Today Bio

journal homepage: www.journals.elsevier.com/materials-today-bio

Application of artificial intelligence in 3D printing physical organ models

Liang Ma a, b, Shijie Yu a, b, Xiaodong Xu a, b, Sidney Moses Amadi c, Jing Zhang a, **,
Zhifei Wang b, *
a
College of Materials Science and Engineering, Zhejiang University of Technology, Hangzhou, 310000, China
b
Zhejiang Provincial People’s Hospital, Hangzhou, Zhejiang, 310000, China
c
International Education College, Zhejiang Chinese Medical University, Hangzhou, Zhejiang, 310000, China

A R T I C L E I N F O A B S T R A C T

Keywords: Artificial intelligence (AI) and 3D printing will become technologies that profoundly impact humanity. 3D
Artificial intelligence printing of patient-specific organ models is expected to replace animal carcasses, providing scenarios that
3D printing simulate the surgical environment for preoperative training and educating patients to propose effective solutions.
Organ models
Due to the complexity of 3D printing manufacturing, it is still used on a small scale in clinical practice, and there
Preoperative training
are problems such as the low resolution of obtaining MRI/CT images, long consumption time, and insufficient
realism. AI has been effectively used in 3D printing as a powerful problem-solving tool. This paper introduces 3D
printed organ models, focusing on the idea of AI application in 3D printed manufacturing of organ models.
Finally, the potential application of AI to 3D-printed organ models is discussed. Based on the synergy between AI
and 3D printing that will benefit organ model manufacturing and facilitate clinical preoperative training in the
medical field, the use of AI in 3D-printed organ model making is expected to become a reality.

1. Introduction independently, as well as perceive and comprehend its surroundings,

working like the human brain, which, in short, can perform all human
Artificial intelligence (AI) is a branch of computer science and en­ tasks. However, no system has yet been developed that fully meets the
gineering that aims to develop intelligent machines that can mimic definition of strong AI, and this technology is still in the research and
human thinking and behavior so that they can perform a range of tasks, exploration phase [9,10]. WAI is a system that can only perform specific
including speech and image recognition, natural language processing, tasks, such as speech recognition, natural language processing, image
autonomous decision-making, and more [1–4]. Its development can be recognition, etc. These systems are usually based on machine learning
traced back to the 1950s and 1960s when John McCarthy first intro­ (ML) and deep learning (DL) techniques. ML is defined as training to
duced the concept at the 1956 Dartmouth Conference [5]. At that time, optimize performance metrics through computer programming using
computer scientists began to study how to make computers think and act experience or example data, where the main technique is neural net­
like humans to achieve intelligent decision-making and operation. In the works [11]. With the increase in computing power and the development
late 1980s, machine learning techniques began to emerge, and methods of large amounts of available data, DL as a branch of ML has shown great
based on statistical models began to occupy an important position in the strength in inference and image processing [12]. Based on these, WAI
field of artificial intelligence. At the same time, as computer computing has been widely used in various fields, including driverless cars, com­
power increased, deep learning techniques began to rise. In the 21st puter graphics, robotics, and spamming, to provide convenience and
century, with the rapid development of cloud computing, big data, the efficiency to human life. In manufacturing, AI technology can provide
Internet of Things, and other technologies, artificial intelligence tech­ onsite information based on sensors and several algorithms to enable
nology has seen substantial growth [6–8]. smart manufacturing, which includes automation and robotics, data
Artificial intelligence technology includes strong and weak artificial analysis and prediction, quality control and defect detection, and pre­
intelligence (WAI). Strong AI, also known as General AI, is designed to dictive maintenance. AI has had a profound impact on manufacturing,
be able to think, understand, learn, and create new knowledge driving the industry towards intelligence and efficiency, and is the

* Corresponding author.
** Corresponding author.
E-mail addresses: [email protected] (J. Zhang), [email protected] (Z. Wang).

https://doi.org/10.1016/j.mtbio.2023.100792
Received 11 August 2023; Received in revised form 1 September 2023; Accepted 8 September 2023
Available online 15 September 2023
2590-0064/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
L. Ma et al. Materials Today Bio 23 (2023) 100792

Fig. 1. The relationship between AI, ML, DL, and the classification of ML applications in 3D printing and the neural networks used by DL. Modified with permission
from Ref. [24]. Copyright 2021 IOPscience.

largest AI investment after the banking, retail, and professional services development of 3D printing technology has accelerated the process of
industries. As a rapidly developing manufacturing technology that product development and marketing and has good potential for the
combines computer-aided design (CAD), computer numerical control manufacture of specific organ models, providing an effective means of
(CNC), mechanical technology, and materials science, 3D printing is a preoperative training for physicians with little surgical experience.
rapid prototyping method that uses CAD data from a computer to Globally, 143 million additional surgical procedures are needed each
transform it into a physical entity [13]. Unlike traditional year in low- and middle-income countries to save lives and prevent
manufacturing techniques, 3D printing creates objects by overlaying disability [18]. However, surgery is a high-risk medical practice that
material layer by layer rather than removing material, which allows 3D requires highly skilled and experienced surgeons to reduce patient
printing to create more complex structures while also increasing pro­ mortality, procedure time, and complications. Surgeons are still limited
ductivity and flexibility. In 1986, Charles Hull developed the first gen­ in simulating clinical procedures due to the lack of training models, and
eration of 3D printing technology (Stereo lithography appearance SLA), 3D printed, manufactured models still have some shortcomings, such as
which is also used in industries such as aerospace and automotive [14]. the time required to process the images, the lack of precision in pro­
Nowadays, 3D printing technology is used in many fields, such as cessing the images, and the errors that occur during the printing process
manufacturing and healthcare [15,16]. It can create items of various that affect the quality of the printed models. In the mid-1990s, there
shapes and sizes, including models of human organs [17]. The were attempts to remotely monitor additive manufacturing (AM)

Fig. 2. Schematic diagram of the different types of direct and indirect printing and direct printing. (A) Direct 3D printing of organ models, 3D printing of casting
molds, and 3D printing of sacrificial material fabrication models. (B)b1 Stereo lithography Appearance (SLA)、b2 Fused Deposition Molding (FDM) 、b3 Inkjet
printing、b4 Digital Light Processing (DLP). Reproduced with permission from Ref. [13]. Copyright 2021 Wiley.

2
L. Ma et al. Materials Today Bio 23 (2023) 100792

Fig. 3. The process of 3D printing organ models. Take 3D printing heart models as an example. Reproduced with permission from Ref. [27]. Copyright 2021 Elsevier.

machines via the web [19,20]. With the rapid development of AI in method is to print the sacrificial material, cover the sacrificial material
recent years, researchers are also innovating ways to leverage AI and ML with a uniform coating, and then melt or dissolve the sacrificial material
integration into AM, such as DL, physical network systems, and neural to obtain the final model, which is usually used to fabricate hollow pipe
networks, designed to automate complex information processing and models. Direct 3D printing is the use of a printer to print the organ model
assist control in helping make rational decisions and self-correction in itself directly on the printing platform without some other extra steps.
product development [8,21]. But there is very little research on the use The mainstream technologies for 3D printing organ models nowadays
of AI combined with 3D printing to produce organ models that can be include stereo lithography appearance (SLA), fused deposition molding
simulated for clinical procedures, while AI is an effective tool for (FDM), inkjet 3D printing, and digital light processing (DLP) (Fig. 2B).
monitoring and controlling the final development object, has a wide Different 3D printing technologies have their advantages and disad­
range of applications in image processing, process monitoring, and in vantages and have different areas of application.
situ correction, has been used to plan 3D printing processes, and has The mechanism of SLA is that the photosensitive liquid resin material
great advantages in achieving this aspect of high-throughput, high-­ generates cations or free radicals under the irradiation of UV light,
quality 3D printed models [22,23] (Fig. 1). shows the relationship be­ which triggers the hybridization polymerization of a liquid resin or
tween AI, ML, DL, and the classification of ML in AM applications. ML photopolymerization of cations and then cures layer by layer to form 3D
algorithms, applications, and platforms can benefit 3D printing tech­ objects [25,26]. SLA is suitable for printing large, high-precision,
nology by enabling the rapid generation of 3D models, the correction of transparent models because it cannot use multiple materials in a sin­
printing errors, and the generation of customized 3D printed models, gle model and, in most cases, requires additional support structures to
among other applications that can bring more convenience and inno­ support the printed part, and the support structures need to be removed
vation. This article talks about the application of AI in printing organ after printing. Examples include heart and blood vessel models for ed­
models and some of the challenges of using AI to print high-quality ucation, training, and flow testing [27].
organ models at high throughput, which allows clinicians to perform FDM is the extrusion of layer-by-layer molten polymer material to
preoperative simulations. form a complete model. This printing technique is simple and can print
prototypes in a variety of sizes, materials, styles, and colors, and it also
2. 3D printed organ models and AI requires the design of support structures when printing some high-
weight models [28]. Although FDM can rapidly print prototypes and
The two manufacturing methods nowadays used for 3D-printed manufacture on demand, it is limited in mass production [29].
organ models are indirect printing and direct printing (Fig. 2A). Indi­ Inkjet printing involves ejecting materials as a specific liquid or ink
rect printing is done by 3D printing an organ mold and then casting the through the nozzle of the printer, and the ejected droplets are deposited
material into the mold to cure it and form the organ model. The other and then cured by UV light or by using in situ chemical reactions to build

3
L. Ma et al. Materials Today Bio 23 (2023) 100792

the object [30,31]. This 3D printing technique can be performed at low the printer for printing. Of course, the software that can perform medical
temperatures and pressures, uses a wide range of materials, and prints in image processing is not limited to Mimics but also includes open-source
a time almost independent of the complexity of the product, but printing software like 3D Slicer, MeVisLab made in Germany, and MITK. Soft­
complex 3D products and hollow structure objects is still limited [32]. ware that can perform CAD processing includes 3-Matic, Magics, Sol­
DLP differs from the point light source of SLA in that DLP uses a idWorks, MeshMixer, etc., which can be selected according to different
projector as a light source to project a layer of a pattern of a model onto a needs.
liquid photosensitive material. The core of the whole system of DLP- 3D printers are used to print models directly after receiving pro­
based 3D printers is the digital micromirror device (DMD) chip, which cessed STL images. Direct 3D printing does not require the demolding
consists of thousands of movable micromirrors that control the intensity and removal of sacrificial materials after the model is formed. However,
of UV light projected onto each layer of image pixels separately, thus to build complex organ and vascular networks, hollow pipelines, and
curing the photosensitive material layer by layer to build the model [26, direct organ-to-organ connections, extrusion-based 3D printing methods
33,34]. As a result, DLP offers high print resolution and the ability to are still limited. To meet the need for printing these complex hollow
print small-sized, high-precision objects [35]. networks and large organs, several 3D printing approaches have been
Using the above-mentioned printing technologies, the printing of developed, such as Freeform Reversible Embedding (FRE) 3D printing
visually interactive organ models and simulated manipulated organ and coaxial extrusion 3D printing [44–46]. FRE 3D printing is equipped
models can be achieved. With the development of 3D printing technol­ with a support tank and a support bath. The printing material can also be
ogy and imaging technology, the combination of these two technologies cured using UV curing, which involves extruding the printing material
has made it possible to produce organ models that are no longer bound into a support bath inside the support tank. The support bath is equiv­
to traditional casting to produce organ models, which can now be alent to Bingham fluid to ensure that the printed model does not
printed directly. The process of 3D printing organ models can be collapse, and after printing, the support bath needs to be removed from
described as (1) image acquisition, (2) image processing, and (3) per­ the top of the model to get the final product [47]. Some larger organs (e.
forming 3D printing (Fig. 3). Overall, the 3D image data of the organ is g., heart, liver, lung, etc.) printed directly on the platform can collapse
obtained using computed tomography (CT), magnetic resonance imag­ due to gravity, and FRE 3D printing can be a good solution to this
ing (MRI), and 3D virtual visualization, the image is then segmented and problem. Coaxial extrusion 3D printing can build some hollow tubular
converted to a format that is imported into CAD software, and finally, structures, which have good applicability in the field of tubular tissue
the 3D image is optimized and sent to a 3D printer for printing [36]. It is engineering [48]. In addition, various types of organ models can be
worth noting that the final organ model obtained is closely related to the printed using DLP by selecting the right material and setting the correct
above steps, especially the first two steps that dominate the good or bad printing parameters. Material jetting allows the printing of abdominal
organ model, so it is especially important to obtain the 3D image data organ models using a variety of materials and precise control of the
and process the image reliably. mechanical properties of the materials [49].
Collecting 3D image data of organs is the key to performing 3D organ The indirect printing method of manufacturing organ models re­
printing, which provides highly accurate information about the organ quires surface treatment and assembly of the printed molds, followed by
structure (e.g., size, morphology, the internal structure of the organ, manual casting, which requires controlling the temperature of the ma­
etc.), which is essential for generating CAD models of the organ so that terial to avoid thermal deformation of the mold and reduce material
the structure and morphology of the organ can be more accurately waste, takes several hours or days to mold, and is difficult to ensure
reproduced during 3D printing. There are many ways to acquire medical accuracy and consistency. When acquiring patient-specific organ
images, including CT, MRI, cone beam computed tomography (CBCT), models, indirect printing undoubtedly increases the manufacturing
positron emission tomography (PET), and ultrasonography (US) process, and the variety of different patient conditions can lead to the
[37–39]. CT is widely used because the image data obtained is easy to need to print a variety of "disposable" molds, thus making the molds
process. wasteful; indirect printing makes it more difficult to produce hollow
The CT scanner consists of an X-ray tube and detector; the X-ray tube tubular models, and the application of materials can lead to uneven
emits X-rays through different parts of the patient’s body, which are coating for human reasons. The use of direct 3D printing technology can
received by the detector and produce a series of image information, then be a good solution to these problems. Current direct 3D printing tech­
the CT scanner converts this image information into digital signals and nology is highly flexible, and models can usually be made within a few
transmits them to the computer, which converts these digital signals into hours, which has led to reduced manufacturing cycles and greatly
two-dimensional images and combines the two-dimensional images into improved efficiency. The exact time depends on the method of 3D
a three-dimensional through computational recovery algorithms; and printing and the size and complexity of the model produced [40]. By
finally, the acquired image information is saved in DICOM format. CT printing organ models directly, we can reduce the waste of materials,
acquisition of 3D image data is an efficient, safe, accurate, and repro­ and with the development of 3D printing technology and innovation of
ducible method for making 3D organ models, which can be scanned at a principles, 3D printing technology can use inexpensive materials to
resolution of 400–600 μm [40]. create organ models, making them cheaper. 3D printed organ models are
After acquiring CT images, the images need to be processed. Mimics more accurate and can be better used in medical education and research,
is a professional 3D medical image processing software mainly used for allowing medical students to use these models for hands-on practice,
model reconstruction, segmentation, restoration, and smoothing oper­ thus improving their skills and knowledge as well as in their clinical
ations while supporting many different image formats and multiple practice. The accuracy of 3D printed organ models can be used for
export formats [41–43]. With its user-friendly interface and easy oper­ medical education and research, allowing medical students to use these
ation, Mimics is suitable for beginners or non-professionals in medical models for hands-on practice, thus improving their skills and knowledge
image processing. Usually, the raw image data obtained after the CT as well as their clinical skills and confidence when performing surgery,
collection of organ images is rough, and there are some irregular bumps achieving better surgical outcomes and thus reducing patient mortality.
or depressions, which can limit the 3D printing process and affect the However, direct 3D printing of organ models can also be limited by
quality of the printed model, so the image data should be processed. several factors. Due to the limitations of printing time and printing
Nowadays, most of the images are reconstructed and segmented with complex structures, it does not allow for high-volume production like
Mimics to extract the organs to be printed, and then the model is further traditional manufacturing. From a material perspective, 3D printed or­
processed with CAD for mesh optimization to meet the printing re­ gans do not fully mimic human organs due to the limited choice of
quirements. At this point, the optimized image data still exists in DICOM materials, and due to the nature of the material itself, the ink used to
format, so the images are converted to standard STL format and sent to perform the printing may clog the printer nozzles at high fluxes, thus

4
L. Ma et al. Materials Today Bio 23 (2023) 100792

Fig. 4. Example of the ML application in 3D printing. (A) An example of supervised learning is where the training model is trained to predict the print suitability of
the ink. (B) An example of super-resolution unsupervised learning is where the model learns from high-resolution images. Low-resolution input data can be generated
automatically without laborious manual annotation. (C) Parameter-optimized reinforcement learning tasks, where desired outcomes, such as resolution and
manufacturing speed, can be used as rewards for training the model. (D) Example methods for deep learning.

malfunctioning and affecting the printing results [50]. From the learning, unsupervised learning, and reinforcement learning [55].
printer’s point of view, the printer must be set up with high accuracy, In supervised learning, the algorithm needs to use labeled training
and the temperature, speed, and other parameters need to be controlled data, i.e., the correct answers to known data, to learn and predict, and as
during the printing process to ensure printing accuracy, which puts high the input data is fed into the model, the model adjusts its weights until it
demands on the printer. From a cost point of view, it takes a long time to fits correctly as part of the cross-validation process to ensure that the
make a high-precision model and to obtain a high-precision model, a model avoids overfitting or underfitting. Some of the methods used in
variety of software is needed to segment and post-process the image, supervised learning include neural networks, plain Bayes, random for­
making the time to make the model longer. There are also laser-based ests, and support vector machines (SVMs), which can help organizations
printing technologies that require the construction of optical printing solve a variety of real-world problems such as image classification,
systems, which require high costs to build these systems [51–53]. speech recognition, medical diagnosis, and some prediction (Fig. 4A).
Finally, 3D-printed organ models need to be validated before they can be illustrates an example of supervised learning, where a large dataset of
widely used in clinical settings. Questions such as whether the materials ink recipes is first collected and labeled, the recipe for the ink to be
are suitable for printing organ models, whether the printer supports the printed is input during training, and the print suitability of the ink is
printing of the materials, and whether the manufactured models still output as a prediction. In unsupervised learning, algorithms are required
need to be verified, which requires rigorous experiments and studies to automatically discover patterns and structures from unlabeled data,
before conclusions can be drawn. and this type of learning is often used for problems such as clustering,
In summary, there is an urgent need to solve the various problems anomaly detection, and dimensionality reduction. The ability of this
encountered in 3D printing organ models to obtain excellent printed approach to discover differences and similarities in information makes it
models. With the continuous development of artificial intelligence, it is ideal for exploratory data analysis as well as image and pattern recog­
now being used in various fields, including 3D printing. To efficiently nition, and it is also used to reduce the number of features in a model
obtain high-quality printed models of organs, printing assisted by AI is a through the process of dimensionality reduction. Algorithms used in
very promising option. unsupervised learning include methods such as principal component
ML is a branch of AI and computer science that focuses on how to use analysis (PCA) and singular value decomposition (SVD), neural net­
computer algorithms and models to allow computers to automatically works, k-means clustering, and probabilistic clustering (Fig. 4B). shows
learn and recognize patterns from data to enable prediction, decision- an example of unsupervised learning where a super-resolution network
making, and model optimization [54]. In ML, a large amount of data is iteratively trained to reconstruct a high-resolution image by blurring
is usually required for training, which can come from various sources, the original image to generate a low-resolution image as input and using
such as sensors, text, images, videos, etc. By feeding this data into ma­ the original high-resolution image as the truth-value output label [12].
chine learning algorithms, the computer can automatically learn the Reinforcement learning is a machine learning model like supervised
patterns and laws in it so that it can predict and classify unknown data. learning, but the algorithm is not trained using sample data and requires
ML algorithms are divided into three main categories: supervised interaction with the environment to learn and make decisions. Through

5
L. Ma et al. Materials Today Bio 23 (2023) 100792

Fig. 5. The three stages of 3D printing organ models and the factors to consider at each stage.

trial-and-error learning, the environment can provide rewards or pen­ before, during, and after printing, and there are many factors to consider
alties to the intelligence, thus helping the machine learning model in this process (Fig. 5), outlines the factors to look for in the 3D printed
maximize rewards and gradually improve its prediction accuracy organ model process.
(Fig. 4C) displays an example of reinforcement learning where, in each
training set, the reinforcement learning agent can observe information
3.1. AI application in pre-printing
about the state of the environment, such as the current printing pa­
rameters and status. By using the value of the key outcome as a reward
AI has penetrated image processing and some predictive models
signal, the intelligence can improve print quality by penalizing behav­
before 3D printing. Generally, after CT scans generate 3D models, AI is
ioral choices that lead to poor outcomes (e.g., the too-high viscosity of
used to help with image segmentation and subsequent optimization of
the ink) while encouraging behavioral choices that lead to positive
CAD models, etc. In terms of prediction, AI works mainly on predictive
outcomes (e.g., appropriate print viscosity). DL is a multi-layer neural
modeling and predicting the print suitability of materials to reduce some
network that learns high-level feature representations of data and en­
trial and error.
ables the recognition and classification of complex patterns (Fig. 4D).
"Deep" machine learning can use labeled datasets (also known as su­
3.1.1. Image processing and pre-process
pervised learning) to inform its algorithms, but it does not necessarily
In 3D printed organ models, the pre-processing step is particularly
need labeled datasets; it can ingest unstructured data in its raw form (e.
important, especially for image reconstruction, which plays a very
g., text or images) and can automatically determine the set of features
crucial role in identifying anatomical structures and obtaining data in
that distinguish between different classes of data, with the ability to
different geometric dimensions. When segmenting and post-processing
handle large-scale data, automatic feature learning, nonlinear models,
the data at the pre-processing stage, the desired results are often not
high accuracy, and generalization capabilities [56,57]. Learning is a
achieved due to objective reasons. However, traditional medical image
very broad topic that has good prospects for application in 3D printing
segmentation methods require a lot of manual effort and expertise, are
[58].
time-consuming and error-prone, and the operator has only a two-
dimensional perception of the image at each point in time [59]. The
3. Application of AI in 3D printing organ models
acquisition of sufficient and high-quality training data by engineers,
including the reorganization of already established data (STLs), unifi­
Successful printing of organ models requires design and optimization
cation and exporting one by one, quality checking one by one, etc., is

Fig. 6. The three stages of 3D printing organ models and the factors to consider at each stage. (A) Image acquisition and processing. Super-resolution is achieved
using unsupervised learning, where the input low-resolution (LR) image is converted to a high-resolution (HR) image after layers. (B) Image Segmentation. Seg­
mentation of high-resolution 2D images is performed to separate the desired 3D tissue by applying a series of thresholds to obtain a 3D surface model. Modified with
permission from Ref. [60]. Copyright 2018 Elsevier.

6
L. Ma et al. Materials Today Bio 23 (2023) 100792

time-consuming and inevitably labor-intensive when the amount of data surfaces, then use AI technology to analyze the morphology and char­
is too large. Young doctors in the clinic need models for surgical acteristics of the surface and optimize the 3D print path through algo­
training, but because of the personal privacy involved in CT images, it is rithms. The model data and some of the calculations take place before
often impossible to use these CT images for batch printing. For these printing, and the algorithms adjust the trajectory of the printer’s nozzles
reasons, it becomes difficult to get common models in clinical and sur­ based on the characteristics of the surface to fit the surface as closely as
gical training. Therefore, processing image data is a critical problem that possible to ensure print quality and accuracy [77]. For example, Zhao
needs to be solved urgently, and AI has been widely utilized in pro­ [78] et al. obtained the conversion between 3D printing coordinates and
cessing medical images (Fig. 6A). showed the training of 3D scanner coordinates by calibrating the external parameters of the
super-resolution neural networks in 3D tissue imaging, using camera in the laser 3D scanning system. They enabled this integrated
high-resolution 3D images as training labels and low-resolution 3D im­ printing system to repair broken gears using a specially designed
ages as input to obtain high-resolution images (Fig. 6B). demonstrates a computational workflow, which opens up the possibility of applying it to
two-stage segmentation method to improve the quality of segmented some organ models that need to be combined.
images, with a roughly defined candidate region as the first stage and Printed products with warped models or reduced surface accuracy
the target organ within the identified region as the second stage to due to material shrinkage and uneven material flow can affect product
complete high-quality segmentation [60]. quality leading to scrap, which can increase the cost of production.
A convolutional neural network (CNN) is an artificial neural network Predictive modeling and compensation methods are used in industrial
specifically designed to process data with grid-like structures, such as production to address product deformation, and such predictive
images, sounds, etc. The core idea is to take the input data (e.g., images) modeling methods are mainly physically based, using finite element
and perform multiple layers of convolution and pooling operations to modeling methods and data-driven methods based on statistics and
extract the features of the data, gradually reduce the size of the data, and machine learning. These methods can effectively predict the mechanical
finally output the prediction results [61]. CNNs are extensively used in behavior of the product during printing, changing factors such as layer
the field of computer vision for tasks such as image classification, target thickness and flow rate to discover the best settings for quality control
detection, and image segmentation. In the medical field, CNNs are also [79–85]. However, the computational complexity as well as the vari­
used in medical image analysis, pathology detection, and medical ability of various scenarios, and the cumbersome modeling process
diagnosis and have achieved many impressive results. Milletari [62] during the simulation have limited its widespread application. Decker
et al. used a fully convolutional neural network (FCN) for the segmen­ [86] et al. proposed a geometric accuracy prediction method based on
tation of 3D volumetric medical images. The researchers conducted triangular mesh shape data to easily build complex 3D models, and the
experiments on several medical image datasets, including the heart, researchers extracted key geometric features and parameters in the 3D
brain, and liver. The method significantly outperformed traditional 2D printing process and combined it with an ML algorithm for prediction
slice-based methods and other 3D segmentation methods in terms of and optimization. The method not only predicts the possible geometric
segmentation accuracy and robustness. Zheng [63] et al. developed an accuracy problems in the 3D printing process but also optimizes and
intelligent image segmentation method based on visual knowledge improves based on the prediction results, thus improving the geometric
representation, which represents medical knowledge as a graphical accuracy and quality of printed products [87–91].summarize some
language, and then uses deep learning techniques to represent this visual ML-based compensation for material properties and geometry
knowledge as weights of a neural network, thus achieving automated prediction.
utilization and integration of medical knowledge. Now, Mimics 25.0 and Some 3D printing methods based on photopolymerization for print­
Mimics Viewer software and AI Assistant plug-in that introduce new ing physical organ models usually require the design of appropriate
deep learning algorithms and new AI algorithms can automatically support structures to support the overhanging parts during printing. AI
identify the contours and internal structure of organs, generate can be applied to the generation and optimization of support structures
high-precision three-dimensional models, and automatically identify to automatically design the best support structure based on the geo­
abnormalities and lesions of organs, and provide corresponding treat­ metric features and mechanical requirements of the model, improving
ment suggestions, helping doctors to diagnose diseases and plan treat­ printing efficiency and reducing post-processing.
ment faster and more accurately, thereby improving treatment effects The use of CNN enables 3D printed organ models in pre-processing to
and patients’ quality of life. And the batch of data collation and export improve the accuracy and reliability of medical images; the use of Open-
has been realized, and the success rate of heart structure reconstruction Loop AI Printing makes printing on complex surfaces a reality, using
of Mimics version 25.0 based on the AI algorithm can be as high as predictive models combined with ML to predict product accuracy and
98.8%. Nvidia’s introduction of Clara also enables AI breakthroughs in make optimizations to reduce trial and error in the printing process, in
medical imaging. It can use AI to assist in the generation of lesion addition, AI in light-cured printing technology based on the support
annotation data and process real-time data streams to ensure that pa­ structure of Automatic generation reduces the burden in pre-processing.
tients are correctly positioned before image acquisition, and integrating Overall, AI improves efficiency for printing high-throughput, high-pre­
Clara in MITK allows for rapid segmentation. Automated reconstruction cision models.
using AI does not require CT image data from patients and enables batch
export of data, which facilitates physicians to obtain 3D-printed organ 3.1.2. Material matching and parameter optimization
models for routine surgical simulation without the need for Choosing the right ink and printing parameters for 3D printing organ
patient-specific CT images and specialized staff to process the images, models is crucial, which is related to the smoothness of the printing
thus reducing time consumption. process and the quality of the printed model. In 3D printing, the nature
In addition, traditional 3D printing technology usually requires of the ink determines the printing parameters, such as printing speed,
printing on flat surfaces, which limits its application scope and effi­ printing temperature, travel distance, dispensing pressure, nozzle
ciency. Using AI can break this limitation by performing 3D printing on diameter, and printing time, while the printing parameters determine
irregular surfaces, making 3D printing technology more intelligent and the degree of damage to the printer and the accuracy of the printed item.
flexible. In 3D printing, organ models often involve printing complex In a variety of material compositions of the ink, due to the ink formu­
and irregular shapes. For example, if part of the printed product is lation of the composition and mechanical properties of the link between
missing due to printing errors during the printing process, compensating the role of ink and ink and some complex non-linear relationships, as
for this part of the missing to reduce material waste, which is often not well as the wide range of materials, the composition of all inks and their
possible with in-situ 3D printing technology. Open-Loop AI Printing can complex formulations are more difficult to study one by one, so deter­
pre-obtain data from models by laser scanning or modeling irregular mining the universal ink printing suitability is still a challenge [92]. At

7
L. Ma et al. Materials Today Bio 23 (2023) 100792

Fig. 7. Example of a 3D printed organ model applying a neural network. (A) The printing parameters are fed into the neural network as training data to predict the
filament diameter and the mechanical properties of the material. (B) Development of flowcharts for ink. Modified with permission from Ref. [94]. Copyright
2022 MDPI.

the same time, choosing the best printing parameters to match the na­ optimized several parameters of the silicone rubber 3D printing process,
ture of the ink requires a lot of effort [93]. Analytical methods such as including print speed, pressure, temperature, and material flow,
ML can be used to identify and predict complex relationships between resulting in better print quality and efficiency and a 2.5-fold increase in
ink formulations and optimize printing parameters to identify and pre­ print speed. Shi [97] et al. used the ML algorithm to optimize the
dict ink performance and print suitability, reducing the cost of ink multi-objective design problem of Drop-on-Demand (DoD) bioprinting,
design and finding the best printing parameters to match ink properties and they obtained high-quality DoD bioprinting results in a short time,
for efficient ink development (Fig. 7A). shows the neural network pre­ increasing the printing speed from 0.88 to 2.08 m s− 1. This approach
dicting print results based on unseen data. The printing parameters are reduced the number of experiments and costs and improved the effi­
fed into the neural network as training data, and then the corresponding ciency of bioprinting technology development. Rojek [98] et al. pro­
results (e.g., mechanical properties) are used to adjust the parameters in posed a model based on artificial neural networks and genetic
the algorithm. The flowchart of the printable ink design is shown in algorithms for computationally optimizing material selection in the 3D
(Fig. 7B). printing process. With this method, the researchers selected the most
Lee [92] et al. developed printable bio-inks of naturally derived suitable material for the hand exoskeleton component, thus improving
biomaterials based on the ML approach, predicting the printability of the performance of the final product. This method is fast and effective,
the inks. Ruberu [95] et al. collected data on the mechanical properties but the process is more complex and sets too many parameters.
of a series of biomaterials, constructed a prediction model using ML and The successful prediction of material properties using AI has reduced
then used this prediction model to predict the mechanical properties of the time consumed by researchers in exploring material formulations to
biomaterials under different 3D printing parameters. The researchers meet printing requirements by algorithmically calculating the optimal
used these prediction results to select the optimal printing parameters printing parameters to match the material properties. In addition, high-
and to experimentally validate that this ML-based optimization method quality products were obtained after printing using AI prediction
could result in high-quality 3D printed biomaterials in a short period. In models, which offers the prospect of using AI for surgical training of 3D
addition, this method can reduce the number of experiments and the printed organ models.
cost of experiments, thus accelerating the development process of bio­
materials. Menon [96] et al. used a method called "hierarchical machine
learning," which divides data into different levels and uses different ML 3.2. AI application in printing
techniques to analyze and optimize them. The researchers successfully
The manufacturing process of 3D printed organ models is a dynamic

Fig. 8. Example of Closed-Loop AI Printing. (A) Closed-loop voltage control framework for the Liquid Metal Jet Printing process. Reproduced with permission from
Ref. [107]. Copyright 2018 Elsevier. (B) Workflow of a closed-loop feedback control algorithm based on a convolutional neural network (CNN) model. Modified with
permission from Ref. [108]. Copyright 2019 Elsevier.

8
L. Ma et al. Materials Today Bio 23 (2023) 100792

one, which also involves several steps, and therefore various problems a remote defect detection method for 3D printing based on computer
can occur at any time, errors and failures can still occur even after the vision and artificial intelligence technology to detect defects such as
prediction of the print suitability of the material and the optimization of beading in real-time. The method uses DL technology to analyze and
the parameters. For example, changes in the environment (e.g., tem­ process the data in the 3D printing process to detect and classify defects
perature changes and vibrations) can affect the behavior of the material, such as beading and accurately detect defects such as beading in 3D
and the mechanical behavior of the printer (e.g., motion errors in the printing process, enabling real-time monitoring and early warning of
drive mechanism and the degree of extrusion) can affect the quality of abnormalities. The method also allows automatic adjustment of 3D
the printed product [105]. During the printing process, the print mate­ printing parameters, thus further improving product quality and effi­
rial loses adhesion to the platform to form defects; uncertainty in the ciency. When defects are generated by individual layers, CNNs can be
material (such as the viscosity and distribution of the ink) can lead to utilized in a closed-loop feedback system to evaluate the printed layers
clogging of the printer nozzles, failure of the support structure leading to [21]. Jin [110] et al. used a DL network-based anomaly monitoring
the collapse of the object, etc. These actions occur when the printer does method to monitor anomalies in the 3D bioprinting process in real-time,
not stop working and can lead to a waste of material and effective including factors such as material fluidity, temperature, and pressure,
equipment running time if no intervention is made [106]. In large-scale thus enabling real-time monitoring and early warning of anomalies. The
printing, the appearance of individual single-layer defects is also diffi­ method can also automatically adjust 3D printing parameters based on
cult to detect, so it is necessary to use AI to identify defects and printing the monitoring results, thus further improving the printing quality and
errors in the printing process and make reasonable choices. Applying ML efficiency.
and CNN to the Closed-Loop system can effectively identify defects that Currently, there are very few studies using AI in the manufacturing
appear during the printing process and automatically correct them. process of 3D printed organ models, but AI based on the 3D printing
Closed-Loop AI Printing is an advanced manufacturing method that process to make parameter adjustments, change the mechanical
combines 3D printing and artificial intelligence technologies to achieve behavior of the printer, and detect defects in the printing process in real
high precision and quality 3D printing by monitoring and adjusting time are widely studied, which has guiding implications for the use of AI
various parameters in the printing process in real-time, using feedback in the manufacturing process of 3D printed organ models used for pre­
loop control to automatically correct errors and adapt to new printing operative training.
requirements [77]. The implementation of Closed-Loop AI Printing re­
quires two main components: a sensor and a control algorithm. The 3.3. AI application in post-printing
sensors are responsible for collecting real-time printing data such as
temperature, pressure, speed, and material flow, while the control al­ Examples of AI applications in the post-processing step of 3D printing
gorithm is responsible for analyzing this data and automatically are rare because the product is already formed at the time of post-
adjusting printing parameters such as print speed, nozzle temperature, processing and cannot be further processed or improved. Therefore,
material flow, etc., according to the target requirements. When changes after printing is completed, quality inspection of the printed product is
in the environment cause deformations in the printed product (e.g., needed to prevent surgeons from training with defective models. In
shrinkage, bending, expansion, etc.), Closed-Loop AI immediately cor­ addition, how to store the products after they are printed is an issue
rects these printing errors, and this feedback loop continues so that worth considering because environmental factors (e.g., temperature,
quality and accuracy are maximized during the printing process PH, air humidity, etc.) can affect the quality of the products in use, e.g.,
(Fig. 8A). shows the closed-loop control framework proposed by Wang printed organ models can become moldy and have poor mechanical
[107] et al. First, a series of dynamic images of the droplet are captured properties, etc. If the printed products are not adequately managed, it is
using a charge-coupled device (CCD) camera, and the droplet features difficult for surgeons to get ideal models to achieve good training results.
and attributes (e.g., satellite, ligament, volume, and speed) are extracted Nevertheless, if defects in printed products appear internally, it is
from the images to quantify the droplet behavior, then a neural network difficult to detect them by visual inspection of staff, and if mass pro­
model is trained to identify the voltage variations, and finally the duction is carried out, it is difficult to avoid the lack of detection of
optimal voltage level is determined by the training model and sent to the defects due to human negligence. Therefore, post-printing processing
voltage regulation system to control the droplet spray behavior. Finally, should summarize the problems affecting the quality of printed models.
the optimal voltage level is determined by the training model and sent to In the manufacturing industry, visual AI and DL used for quality in­
the voltage regulation system to control the droplet injection behavior spection have been reported many times, and relying on the resources
(Fig. 8B). indicates the application of CNN in a closed-loop printing currently available to solve quality problems, AI is the most appropriate
system. A residual network model trained using an image dataset has and advanced technology.
been successfully integrated into a fused deposition modeling 3D printer As mentioned earlier, CNN can be used for the recognition and
equipped with a webcam to monitor the printing process in real-time. training of images, and it is used in a wide range of applications and has
The model can extract spatial hierarchical features from images and been used many times for defect detection in industrial production. Lee
identify the quality of print states, including three states such as good [121] et al. used data enhancement techniques to process PC (Precast
quality, under-extrusion, and over-extrusion. The trained model is then Concrete) component images and then trained and recognized the pro­
deployed into a closed-loop printing system that uses real-time video cessed images with CNN to achieve crack detection and quality in­
streaming for extrusion problem identification and adjusts the material spection of PC components. This technique can greatly improve the
flow rate accordingly to ensure print quality control and improvement. accuracy and efficiency of detection and provides better support for
Shi [109] et al. proposed a machine learning-based method to con­ construction and quality assurance of OSC (Off-Site Construction) pro­
trol the flow and position of dribble cell printing. They used a high-speed jects. In addition, CNNs are used for quality inspection. Wang [122]
camera to capture images of each droplet during the printing process et al. proposed a DCNN-based coffee bean detection technique to ach­
and compared them with the predicted flow rate and position. ML ieve the detection and classification of coffee beans with high accuracy
techniques are then used to analyze and adjust parameters during and efficiency. Sundaram [123] et al. constructed a DL algorithm based
printing, including printing speed, fluid viscosity, and pressure, to on CNN architecture and proposed an AI-based Smart Quality Inspection
improve the accuracy and stability of printing. Traditional 3D printing (SQI) method for manufacturing, which achieves 99.86% accuracy in
defect detection methods usually require manual operation, which is not inspecting cast products but is limited in data collection. Jung [124]
efficient; therefore, the use of computer vision and artificial intelligence et al. proposed an AI-based ACED (Anomaly Candidate Extraction and
technology to achieve 3D printing remote defect detection is important Detection) detection method, researchers constructed an effective defect
to improve the quality of products. Paraskevoudis [106] et al. proposed detection system based on DNNs to detect the surface of metal casting

9
L. Ma et al. Materials Today Bio 23 (2023) 100792

Fig. 9. Future perspectives on the use of AI for 3D printing of organ models. Modified with permission from Ref. [24]. Copyright 2021 PubMed Central.

products and quickly and accurately detect the anomalies in metal combination of 3D-printed organ models and AI technology for simu­
casting products, but when the defects are very small, the A large lating and predicting surgical procedures is equally promising, helping
number of input pixels can be overfitted to reduce the prediction ability, doctors to assess surgical risks preoperatively, choose the best surgical
for example, some small defects such as scratches and dust that are not strategy, and reduce accidents and complications during surgery. The
artificially considered quality problems can also be recognized by the AI combination of AI and 3D printing will bring more efficient, accurate,
inspection system leading to false positive detection, which requires and personalized preoperative training models to the medical commu­
classification of the data set. nity, improving the quality of surgery and patient care outcomes. As
In the 3D-printed organ surgery model, CNN is used in the post- technology continues to develop and innovate, it is believed that this
processing stage to detect the quality of the printed products and clever combination will bring more breakthroughs and advances to the
reject those with quality problems. Computer vision technology is used medical field (Fig. 9). illustrates the application of AI in the whole
to process visual information such as video and images to monitor the process of 3D organ model printing, i.e., before, during, and after
environment and make timely adjustments when the environment printing.
changes, monitor the printed products, find some problems with the Unfortunately, manufacturing high-quality 3D models requires ac­
models in time, and optimize the storage environment of the models to curate medical image processing and large amounts of data, and
ensure the quality of the products. With the whole process, the use of AI currently, access to specific models requires access to patient CT images,
can fully escort surgeons to get high-quality models and promote the which is very limited, and access to patient CT images involves patient
generation of 3D-printed organ models efficiently. privacy, which requires increased regulation of digital medical tech­
nologies to protect patient privacy. ML-based printing requires the
4. Prospects and challenges collection of large amounts of data to train models to improve prediction
accuracy, and nowadays, there is a wide variety of access to data, so
AI is a powerful tool that identifies and integrates large amounts of there is a need for reliable and available data collection, storage, and
data and uses that data to calculate optimal solutions. 3D printing is a sharing, which requires the creation of databases and storage and data
rapid prototyping method that can be combined with AI to produce disclosure, and it is good to see that with the expansion of open source,
high-precision products quickly and accurately. The use of AI has some data can be publicly accessible [94]. The process of 3D printing
accelerated the development of 3D printing inks and reduced the trial involves the interaction of multiple factors, such as temperature, speed,
and error of researchers in the adaptation of inks for printing. With the and material, which makes it difficult for AI algorithms to accurately
rapid development of 3D printing in recent years, 3D-printed organ capture each of these influences, and most available techniques gener­
models are gradually replacing live animals and cadavers for the pre­ ally rely on surface and optical imaging of the material, which can be
operative training of young physicians, which offers great potential for difficult to obtain if defects occur in deeper and more obscure places,
3D printing of organ models for preoperative training. However, the lack and for small models, these defects can affect the effectiveness of the
of specific models, the lack of fidelity of the printed models, and the high simulation. For large models, small surface scratches, stains, and other
printing cost limit the large-scale printing of organ models, and the "defects" that do not affect quality can cause over-recognition by the AI
rational use of AI can make 3D printed organ models benefit from it. In and therefore require classification of defects, which requires more data
the future, AI technology will enable automatic processing and analysis collection. Although AI can improve the efficiency of 3D printing and
of large amounts of medical image data, rapid generation of accurate 3D save some time on 3D printing research, the development and deploy­
models, maximum optimization of printing parameters, control of ment of AI algorithms require a lot of effort, and for small-scale printing,
printer printing behavior, printing on complex surfaces, detection of these costs are too high.
product anomalies, and improvement of printed product quality in an
efficient manner, which will better help young doctors improve their
operational skills and proficiency in surgery. Furthermore, the

10
L. Ma et al. Materials Today Bio 23 (2023) 100792

Table 1
AI algorithms for image processing and segmentation.
Purpose Method Inputs Outputs References

Achieve precise image super- deep convolutional network, Generative adversarial network LR (Low- HR (High-resolution) images [64–66]
resolution (GANs), Deep 3D convolutional networks resolution)
images
Conversion of clinical brain MRI SynthSR, the method relies on CNN LR composite 1 mm isotropic magnetization [67]
scans to high-resolution T1 scanning prepared - rapid gradient echo
scans (MPRAGE)scan
Resolving the low-resolution Enhanced fast super-resolution convolutional neural network LR CXR images HR images [68]
characteristics of CXR images (EFSRCNN), CNN-based
Image Segmentation ANN (Artificial Neural Network), A layered 3D U-Net architecture High duration or Low duration or high accuracy [69–75]
deep learning algorithm, Conjugate full convolutional network low accuracy
(CFCN), FCN (Fully convolutional network), Deep multiscale
convolutional neural network (DMCNN), 3D convolutional neural
network (3D CNN)
Three-dimensional liver 3D CNN High Duration Low Duration [76]
reconstruction automation,
vascular extraction

data collection, storage, processing, and analysis. The open-source

Table 2 extension allows some data to be publicly accessible, which facilitates
AI algorithms for optimization and prediction.
the accumulation of data and provides the impetus for DL. Finally, AI-
Process Purpose Method References based 3D printing of organ models ensures high efficiency and quality
Pneumatic Screen and optimize artificial intelligence- [99] of printed products comprehensively, from efficient and high-
extrusion bioprinting assisted high- throughput image processing, predicting the print suitability of mate­
bioprinting conditions throughput rials, optimizing printing parameters, intelligently monitoring the
printing-condition-
screening system (AI-
printing process, and monitoring and predicting the finished product
HTPCSS) promptly, which provides a path for young surgeons with rudimentary
Extrusion-based Parameter machine learning (ML), [100–102] skills to become highly skilled "veterans.” Therefore, the application of
printing optimization, quality deep learning (DL), AI to 3D printing organ models for preoperative training is an area worth
control, and anomaly ANN, Whale
investigating (see Tables 1–3).
detection optimization algorithm
(WOA)
Inkjet-Based Predict droplet ensemble learning [103] Author contributions
Bioprinting velocity and volume
Powder Bed Parameter Finite element Method [104] M.L drafted the manuscript, and W.Z.F. and J.Z. defined the focus of
Fusion (PBF) optimization (FEM), ML
Freeform Optimize materials Hierarchical machine [96]
the review. Y.S.J. modified the manuscript. Sidney provided language
Reversible and formulations learning (HML) assistance for the manuscript. X.X.D. modified it into the final edition.
Embedding All authors have read and agreed to the published version of the
(FRE) manuscript.

Funding
Table 3
AI algorithms for proto-monitoring and error correction. This work was supported by the Clinical Applied Research Program
Process Purpose Method References of the Department of Health of Zhejiang Province (C-2022-W1062) and
the Key Research and Development Special Project of 2022 Autonomous
PBF In-situ process monitoring, part ML, DCNN, [111–115]
quality identification, ANN Region (20222101517).
intelligent correction
– Defect detection, intelligent Closed-loop [116] Declaration of competing interest
adjustment, and correction ML algorithm
FDM In-situ defect correction DL, CNN [21,117,
118]
The authors declare that they have no known competing financial
selective laser In-situ monitoring Deep belief [119] interests or personal relationships that could have appeared to influence
melting network the work reported in this paper.
(SLM) (DBN)
Surface defect detection ANN [120]
Data availability
–

5. Conclusions No data was used for the research described in the article.

3D printing of organ models for preoperative training can help Acknowledgments

inexperienced young surgeons improve their surgical technique to
reduce risks during surgery and allow patients to better understand the This work was supported by the Clinical Applied Research Program
condition, but the quality of the printed products is affected by factors of the Department of Health of Zhejiang Province (C-2022-W1062) and
such as the long printing process, the low resolution of the acquired the Key Research and Development Special Project of the Autonomous
images, printing parameters that are not optimal, printing errors that Region in 2022 (20222101517). The authors would like to thank Sidney
occur during printing, and improper storage of the printed products. for his language assistance during the writing process.
Fortunately, with the rapid development of AI technology, some algo­
rithms applied to 3D printing can perfectly solve these problems through

11
L. Ma et al. Materials Today Bio 23 (2023) 100792

References [27] D.D. Wang, Z. Qian, M. Vukicevic, S. Engelhardt, A. Kheradvar, C. Zhang, S.

H. Little, J. Verjans, D. Comaniciu, W.W. O’Neill, M.A. Vannan, 3D printing,
computational modeling, and artificial intelligence for structural heart disease,
[1] C. Hyunjin, A study on the change of manufacturing design process due to the
JACC (J. Am. Coll. Cardiol.): Cardiovascul. Imaging 14 (2021) 41–60, https://doi.
development of A.I design and 3D printing, IOP Conf. Ser. Mater. Sci. Eng. 727
org/10.1016/j.jcmg.2019.12.022.
(2020), 012010, https://doi.org/10.1088/1757-899X/727/1/012010.
[28] B. Berman, 3-D printing: the new industrial revolution, Bus. Horiz. 55 (2012)
[2] W. Rawat, Z. Wang, Deep convolutional neural networks for image classification:
155–162, https://doi.org/10.1016/j.bushor.2011.11.003.
a comprehensive review, Neural Comput. 29 (2017) 2352–2449, https://doi.org/
[29] A. Cano-Vicent, M.M. Tambuwala, SkS. Hassan, D. Barh, A.A.A. Aljabali,
10.1162/neco_a_00990.
M. Birkett, A. Arjunan, Á. Serrano-Aroca, Fused deposition modelling: current
[3] E.V. Garcia, Integrating artificial intelligence and natural language processing for
status, methodology, applications and future prospects, Addit. Manuf. 47 (2021),
computer-assisted reporting and report understanding in nuclear cardiology,
102378, https://doi.org/10.1016/j.addma.2021.102378.
J. Nucl. Cardiol. 30 (2023) 1180–1190, https://doi.org/10.1007/s12350-022-
[30] Z. Jin, Y. Li, K. Yu, L. Liu, J. Fu, X. Yao, A. Zhang, Y. He, 3D printing of physical
02996-5.
organ models: recent developments and challenges, Adv. Sci. 8 (2021), 2101394,
[4] J.H. Shin, J. Kwon, J.U. Kim, H. Ryu, J. Ok, S. Joon Kwon, H. Park, T. Kim,
https://doi.org/10.1002/advs.202101394.
Wearable EEG electronics for a Brain–AI Closed-Loop System to enhance
[31] Q. Qian, J.H. Kamps, B. Price, H. Gu, R. Wildman, R. Hague, B. Begines, C. Tuck,
autonomous machine decision-making, Npj Flex Electron 6 (2022) 32, https://
3D reactive inkjet printing of bisphenol A-polycarbonate, Addit. Manuf. 54
doi.org/10.1038/s41528-022-00164-w.
(2022), 102745, https://doi.org/10.1016/j.addma.2022.102745.
[5] J. McCarthy, The inversion of functions defined by turing machines, in: C.
[32] K. Zub, S. Hoeppener, U.S. Schubert, Inkjet printing and 3D printing strategies for
E. Shannon, J. McCarthy (Eds.), Automata Studies. (AM-34), Princeton University
biosensing, analytical, and diagnostic applications, Adv. Mater. 34 (2022),
Press, 1956, pp. 177–182, https://doi.org/10.1515/9781400882618-009.
2105015, https://doi.org/10.1002/adma.202105015.
[6] J. Robertson, J. Fossaceca, K. Bennett, A cloud-based computing framework for
[33] J. Gong, Y. Qian, K. Lu, Z. Zhu, L. Siow, C. Zhang, S. Zhou, T. Gu, J. Yin, M. Yu,
artificial intelligence innovation in support of multidomain operations, IEEE
H. Wang, H. Yang, Digital light processing (DLP) in tissue engineering: from
Trans. Eng. Manag. 69 (2022) 3913–3922, https://doi.org/10.1109/
promise to reality, and perspectives, Biomed. Mater. 17 (2022), 062004, https://
TEM.2021.3088382.
doi.org/10.1088/1748-605X/ac96ba.
[7] J. Zhang, D. Tao, Empowering Things with intelligence: a survey of the progress,
[34] J. Wallace, M.O. Wang, P. Thompson, M. Busso, V. Belle, N. Mammoser, K. Kim,
challenges, and opportunities in artificial intelligence of Things, IEEE Internet
J.P. Fisher, A. Siblani, Y. Xu, J.F. Welter, D.P. Lennon, J. Sun, A.I. Caplan,
Things J. 8 (2021) 7789–7817, https://doi.org/10.1109/JIOT.2020.3039359.
D. Dean, Validating continuous digital light processing (cDLP) additive
[8] Y. Wang, P. Zheng, T. Peng, H. Yang, J. Zou, Smart additive manufacturing:
manufacturing accuracy and tissue engineering utility of a dye-initiator package,
current artificial intelligence-enabled methods and future perspectives, Sci. China
Biofabrication 6 (2014), 015003, https://doi.org/10.1088/1758-5082/6/1/
Technol. Sci. 63 (2020) 1600–1611, https://doi.org/10.1007/s11431-020-1581-
015003.
2.
[35] M.M. Rahman, Statistical Analysis of the Digital Micromirror Devices Hinge Sag
[9] H. Lu, Y. Li, M. Chen, H. Kim, S. Serikawa, Brain intelligence: go beyond artificial
Phenomenon, (n.d.).
intelligence, Mobile Network. Appl. 23 (2018) 368–375, https://doi.org/
[36] A. Ganguli, G.J. Pagan-Diaz, L. Grant, C. Cvetkovic, M. Bramlet, J. Vozenilek,
10.1007/s11036-017-0932-8.
T. Kesavadas, R. Bashir, 3D printing for preoperative planning and surgical
[10] P. Demasi, J.L. Szwarcfiter, A.J.O. Cruz, A theoretical framework to formalize
training: a review, Biomed. Microdevices 20 (2018) 65, https://doi.org/10.1007/
AGI-hard problems, in: Proceedings of the 3d Conference on Artificial General
s10544-018-0301-9.
Intelligence (AGI-10), Atlantis Press, Lugano, Switzerland, 2010, https://doi.org/
[37] M.D. Seemann, J. Gaa, Cardiac metastasis: visualization with positron emission
10.2991/agi.2010.14.
tomography, computed tomography, magnetic resonance imaging, positron
[11] E. Alpaydin, Introduction to Machine Learning, third ed., The MIT Press,
emission tomography/computed tomography, and positron emission
Cambridge, Massachusetts, 2014.
tomography/magnetic resonance imaging, Circulation 112 (2005), https://doi.
[12] W.L. Ng, A. Chan, Y.S. Ong, C.K. Chua, Deep learning for fabrication and
org/10.1161/CIRCULATIONAHA.105.528752.
maturation of 3D bioprinted tissues and organs, Virtual Phys. Prototyp. 15 (2020)
[38] J. Timmer, M. Grass, T. Kohler, R. Proksa, The n-PI-method for helical cone-beam
340–358, https://doi.org/10.1080/17452759.2020.1771741.
CT, IEEE Trans. Med. Imag. 19 (2000) 848–863, https://doi.org/10.1109/
[13] Z. Jin, Y. Li, K. Yu, L. Liu, J. Fu, X. Yao, A. Zhang, Y. He, 3D printing of physical
42.887834.
organ models: recent developments and challenges, Adv. Sci. 8 (2021), 2101394,
[39] I.R.S. Lucena, M.F. Chedid, P.S. Isolan, E.E. Takamatu, R.A. Lucena, F.H. Feier, Á.
https://doi.org/10.1002/advs.202101394.
P.A. Furtado, J.C.S. de Fraga, A comparison between ultrasonography and single-
[14] Y.E. Kalay, Dr -Ing. M. Eigner, "Rechnereinsatz in der mechanischen, (n.d.).
phase computed tomography for preoperative assessment of solid abdominal
[15] R. Bogue, 3D printing: the dawn of a new era in manufacturing? Assemb. Autom.
tumors in children, J. Pediatr. 99 (2023) 17–22, https://doi.org/10.1016/j.
33 (2013) 307–311, https://doi.org/10.1108/AA-06-2013-055.
jped.2022.05.003.
[16] M. Ramola, V. Yadav, R. Jain, On the adoption of additive manufacturing in
[40] F. Rengier, A. Mehndiratta, H. von Tengg-Kobligk, C.M. Zechmann,
healthcare: a literature review, JMTM 30 (2019) 48–69, https://doi.org/
R. Unterhinninghofen, H.-U. Kauczor, F.L. Giesel, 3D printing based on imaging
10.1108/JMTM-03-2018-0094.
data: review of medical applications, Int. J. CARS 5 (2010) 335–341, https://doi.
[17] D. Liu, P. Jiang, Y. Wang, Y. Lu, J. Wu, X. Xu, Z. Ji, C. Sun, X. Wang, W. Liu,
org/10.1007/s11548-010-0476-x.
Engineering tridimensional hydrogel tissue and organ phantoms with tunable
[41] P.J. Reynisson, M. Scali, E. Smistad, E.F. Hofstad, H.O. Leira, F. Lindseth, T.
springiness, Adv. Funct. Mater. (2023), 2214885, https://doi.org/10.1002/
A. Nagelhus Hernes, T. Amundsen, H. Sorger, T. Langø, Airway segmentation and
adfm.202214885.
centerline extraction from thoracic CT – comparison of a new method to state of
[18] J.G. Meara, A.J.M. Leather, L. Hagander, B.C. Alkire, N. Alonso, E.A. Ameh, S.
the art commercialized methods, PLoS One 10 (2015), e0144282, https://doi.
W. Bickler, L. Conteh, A.J. Dare, J. Davies, E.D. Mérisier, S. El-Halabi, P.
org/10.1371/journal.pone.0144282.
E. Farmer, A. Gawande, R. Gillies, S.L.M. Greenberg, C.E. Grimes, R.L. Gruen, E.
[42] G. Luo, Y. Zhang, X. Wang, S. Chen, D. Li, M. Yu, Individualized 3D printing-
A. Ismail, T.B. Kamara, C. Lavy, G. Lundeg, N.C. Mkandawire, N.P. Raykar, J.
assisted repair and reconstruction of neoplastic bone defects at irregular bone
N. Riesel, E. Rodas, J. Rose, N. Roy, M.G. Shrime, R. Sullivan, S. Verguet,
sites: exploration and practice in the treatment of scapular aneurysmal bone cysts,
D. Watters, T.G. Weiser, I.H. Wilson, G. Yamey, W. Yip, Global Surgery 2030:
BMC Muscoskel. Disord. 22 (2021) 984, https://doi.org/10.1186/s12891-021-
evidence and solutions for achieving health, welfare, and economic development,
04859-5.
Lancet 386 (2015) 569–624, https://doi.org/10.1016/S0140-6736(15)60160-X.
[43] L. He, Y. Xu, D. Duan, L. Ouyang, The anterior talofibular ligament: a thin-slice
[19] M.J. Bailey, Tele-Manufacturing: rapid prototyping on the Internet, IEEE Comput.
three-dimensional magnetic resonance imaging study, Foot Ankle Surg. 28 (2022)
Grap. Appl. 15 (1995) 20–26, https://doi.org/10.1109/38.469519.
1202–1209, https://doi.org/10.1016/j.fas.2021.11.011.
[20] R.C. Luo, Jyh Hwa Tzou, Y.C. Chang, Desktop rapid prototyping system with
[44] Y. Fang, Y. Guo, B. Wu, Z. Liu, M. Ye, Y. Xu, M. Ji, L. Chen, B. Lu, K. Nie, Z. Wang,
supervisory control and monitoring through Internet, IEEE ASME Trans.
J. Luo, T. Zhang, W. Sun, Z. Xiong, Expanding embedded 3D bioprinting
Mechatron. 6 (2001) 399–409, https://doi.org/10.1109/3516.974853.
capability for engineering complex organs with Freeform vascular networks, Adv.
[21] Z. Jin, Z. Zhang, G.X. Gu, Autonomous in-situ correction of fused deposition
Mater. 35 (2023), 2205082, https://doi.org/10.1002/adma.202205082.
modeling printers using computer vision and deep learning, Manufact. Letters 22
[45] W. Sun, J.W. Tashman, D.J. Shiwarski, A.W. Feinberg, V.A. Webster-Wood, Long-
(2019) 11–15, https://doi.org/10.1016/j.mfglet.2019.09.005.
fiber embedded hydrogel 3D printing for structural reinforcement, ACS Biomater.
[22] I. Rojek, Intelligent System Supporting Technological Process Planning for
Sci. Eng. 8 (2022) 303–313, https://doi.org/10.1021/acsbiomaterials.1c00908.
Machining and 3D Printing, D Printing., 2021.
[46] H. Ramezani, L. Zhou, L. Shao, Y. He, Coaxial 3D bioprinting of organ prototyps
[23] H. Lee, Engineering in vitro models: bioprinting of organoids with artificial
from nutrients delivery to vascularization, J. Zhejiang Univ. - Sci. A. 21 (2020)
intelligence, Cyborg Bionic Syst 4 (2023), 0018, https://doi.org/10.34133/
859–875, https://doi.org/10.1631/jzus.A2000261.
cbsystems.0018.
[47] C.A. Wu, Y. Zhu, A. Venkatesh, C.J. Stark, S.H. Lee, Y.J. Woo, Optimization of
[24] L. Jannesari Ladani, Applications of artificial intelligence and machine learning in
Freeform reversible embedding of suspended hydrogel microspheres for
metal additive manufacturing, J. Phys. Mater. 4 (2021), 042009, https://doi.org/
substantially improved three-dimensional bioprinting capabilities, tissue
10.1088/2515-7639/ac2791.
engineering Part C, Methods 29 (2023) 85–94, https://doi.org/10.1089/ten.
[25] S. Deshmane, P. Kendre, H. Mahajan, S. Jain, Stereolithography 3D Printing
tec.2022.0214.
Technology in Pharmaceuticals: a Review, Drug Development and Industrial
[48] Y. Zhang, S. Li, X. Zhou, L. Dong, Q. Meng, J. Yu, Preparation of a cellulosic
Pharmacy. (n.d.).
photosensitive hydrogel for tubular tissue engineering, ACS Appl. Bio Mater. 6
[26] H. Quan, T. Zhang, H. Xu, S. Luo, J. Nie, X. Zhu, Photo-curing 3D printing
(2023) 848–856, https://doi.org/10.1021/acsabm.2c01003.
technique and its challenges, Bioact. Mater. 5 (2020) 110–115, https://doi.org/
10.1016/j.bioactmat.2019.12.003.

12
L. Ma et al. Materials Today Bio 23 (2023) 100792

[49] A. Pietrabissa, S. Marconi, E. Negrello, V. Mauri, A. Peri, L. Pugliese, E. learning for automatic mandibular segmentation in cone-beam computed
M. Marone, F. Auricchio, An overview on 3D printing for abdominal surgery, tomography, J. Dent. 114 (2021), 103786, https://doi.org/10.1016/j.
Surg. Endosc. 34 (2020) 1–13, https://doi.org/10.1007/s00464-019-07155-5. jdent.2021.103786.
[50] J. Jang, J.Y. Park, G. Gao, D.-W. Cho, Biomaterials-based 3D cell printing for [74] R. Wang, S. Cao, K. Ma, Y. Zheng, D. Meng, Pairwise learning for medical image
next-generation therapeutics and diagnostics, Biomaterials 156 (2018) 88–106, segmentation, Med. Image Anal. 67 (2021), 101876, https://doi.org/10.1016/j.
https://doi.org/10.1016/j.biomaterials.2017.11.030. media.2020.101876.
[51] C. Lupulescu, Z. Sun, A systematic review of the clinical value and applications of [75] S. Niyas, S. Chethana Vaisali, I. Show, T.G. Chandrika, S. Vinayagamani,
three-dimensional printing in renal surgery, JCM 8 (2019) 990, https://doi.org/ C. Kesavadas, J. Rajan, Segmentation of focal cortical dysplasia lesions from
10.3390/jcm8070990. magnetic resonance images using 3D convolutional neural networks, Biomed.
[52] V. Keriquel, H. Oliveira, M. Rémy, S. Ziane, S. Delmond, B. Rousseau, S. Rey, Signal Process Control 70 (2021), 102951, https://doi.org/10.1016/j.
S. Catros, J. Amédée, F. Guillemot, J.-C. Fricain, In situ printing of mesenchymal bspc.2021.102951.
stromal cells, by laser-assisted bioprinting, for in vivo bone regeneration [76] T. Takamoto, D. Ban, S. Nara, T. Mizui, D. Nagashima, M. Esaki, K. Shimada,
applications, Sci. Rep. 7 (2017) 1778, https://doi.org/10.1038/s41598-017- Automated three-dimensional liver reconstruction with artificial intelligence for
01914-x. virtual hepatectomy, J. Gastrointest. Surg. 26 (2022) 2119–2127, https://doi.
[53] F. Guillemot, A. Souquet, S. Catros, B. Guillotin, Laser-assisted cell printing: org/10.1007/s11605-022-05415-9.
principle, physical parameters versus cell fate and perspectives in tissue [77] Z. Zhu, D.W.H. Ng, H.S. Park, M.C. McAlpine, 3D-printed multifunctional
engineering, Nanomedicine 5 (2010) 507–515, https://doi.org/10.2217/ materials enabled by artificial-intelligence-assisted fabrication technologies, Nat.
nnm.10.14. Rev. Mater. 6 (2020) 27–47, https://doi.org/10.1038/s41578-020-00235-2.
[54] L. Meng, B. McWilliams, W. Jarosinski, H.-Y. Park, Y.-G. Jung, J. Lee, J. Zhang, [78] X. Zhao, Y. Pan, C. Zhou, Y. Chen, C.C.L. Wang, An integrated CNC accumulation
Machine learning in additive manufacturing: a review, JOM 72 (2020) system for automatic building-around-inserts, J. Manuf. Process. 15 (2013)
2363–2377, https://doi.org/10.1007/s11837-020-04155-y. 432–443, https://doi.org/10.1016/j.jmapro.2013.05.009.
[55] C. Yu, J. Jiang, A perspective on using machine learning in 3D bioprinting, Int. J. [79] D. Pal, N. Patil, K. Zeng, B. Stucker, An integrated approach to additive
Bioprint 6 (2020), https://doi.org/10.18063/ijb.v6i1.253. manufacturing simulations using physics based, coupled multiscale process
[56] Y. LeCun, Y. Bengio, G. Hinton, Deep learning, Nature 521 (2015) 436–444, modeling, J. Manuf. Sci. Eng. 136 (2014), 061022, https://doi.org/10.1115/
https://doi.org/10.1038/nature14539. 1.4028580.
[57] S. Dong, P. Wang, K. Abbas, A survey on deep learning and its applications, [80] A. Cattenone, S. Morganti, G. Alaimo, F. Auricchio, Finite element analysis of
Comput. Sci. Rev. 40 (2021), 100379, https://doi.org/10.1016/j. additive manufacturing based on fused deposition modeling: distortions
cosrev.2021.100379. prediction and comparison with experimental data, J. Manuf. Sci. Eng. 141
[58] D. Verma, Y. Dong, M. Sharma, A.K. Chaudhary, Advanced Processing of 3D (2019), 011010, https://doi.org/10.1115/1.4041626.
Printed Biocomposite Materials Using Artificial Intelligence, Materials and [81] A. Hussein, L. Hao, C. Yan, R. Everson, Finite element simulation of the
Manufacturing Processes. (n.d.). temperature and stress fields in single layers built without-support in selective
[59] P. Tavousi, Z. Shahbazi, S. Shahbazmohamadi, AI-Based brain image laser melting, Mater. Des. 52 (2013) 638–647, https://doi.org/10.1016/j.
segmentation using synthesized data, Microsc. Microanal. 26 (2020) 694–697, matdes.2013.05.070 (1980-2015).
https://doi.org/10.1017/S143192762001555X. [82] Z. Li, Z. Zhang, J. Shi, D. Wu, Prediction of surface roughness in extrusion-based
[60] H.R. Roth, H. Oda, X. Zhou, N. Shimizu, Y. Yang, Y. Hayashi, M. Oda, additive manufacturing with machine learning, Robot. Comput. Integrated
M. Fujiwara, K. Misawa, K. Mori, An application of cascaded 3D fully Manuf. 57 (2019) 488–495, https://doi.org/10.1016/j.rcim.2019.01.004.
convolutional networks for medical image segmentation, Comput. Med. Imag. [83] Y. Tu, A. Hassan, J.A. Arrieta-Escobar, U. Khaleeq Uz Zaman, F. Xue, S. Ali,
Graph. 66 (2018) 90–99, https://doi.org/10.1016/j.compmedimag.2018.03.001. G. Yang, Predictive modeling of extruded filament in the air for bioink in direct
[61] S. Minaee, Y.Y. Boykov, F. Porikli, A.J. Plaza, N. Kehtarnavaz, D. Terzopoulos, ink writing using numerical simulation, Procedia CIRP 112 (2022) 394–399,
Image segmentation using deep learning: a survey, IEEE Trans. Pattern Anal. https://doi.org/10.1016/j.procir.2022.09.026.
Mach. Intell. (2021), https://doi.org/10.1109/TPAMI.2021.3059968, 1–1. [84] Z. Zhang, L. Poudel, Z. Sha, W. Zhou, D. Wu, Data-driven predictive modeling of
[62] F. Milletari, N. Navab, S.-A. Ahmadi, V-Net, Fully convolutional neural networks tensile behavior of parts fabricated by cooperative 3D printing, J. Comput. Inf.
for volumetric medical image segmentation, in: 2016 Fourth International Sci. Eng. 20 (2020), 021002, https://doi.org/10.1115/1.4045290.
Conference on 3D Vision (3DV), IEEE, Stanford, CA, USA, 2016, pp. 565–571, [85] L. Cheng, A. Wang, F. Tsung, A prediction and compensation scheme for in-plane
https://doi.org/10.1109/3DV.2016.79. shape deviation of additive manufacturing with information on process
[63] N. Zheng, J. Liu, Visual knowledge representation and intelligent image parameters, IISE Transactions 50 (2018) 394–406, https://doi.org/10.1080/
segmentation, J. Comput. Sci. Technol. 7 (1992) 219–225, https://doi.org/ 24725854.2017.1402224.
10.1007/BF02946571. [86] N. Decker, M. Lyu, Y. Wang, Q. Huang, Geometric accuracy prediction and
[64] J. Kim, J.K. Lee, K.M. Lee, Accurate image super-resolution using very deep improvement for additive manufacturing using triangular mesh shape data,
convolutional networks, in: 2016 IEEE Conference on Computer Vision and J. Manuf. Sci. Eng. 143 (2021), 061006, https://doi.org/10.1115/1.4049089.
Pattern Recognition (CVPR), IEEE, Las Vegas, NV, USA, 2016, pp. 1646–1654, [87] G.X. Gu, C.-T. Chen, D.J. Richmond, M.J. Buehler, Bioinspired hierarchical
https://doi.org/10.1109/CVPR.2016.182. composite design using machine learning: simulation, additive manufacturing,
[65] I. Sanchez, V. Vilaplana, Brain MRI Super-resolution Using 3D Generative and experiment, Mater. Horiz. 5 (2018) 939–945, https://doi.org/10.1039/
Adversarial Networks, 2018. http://arxiv.org/abs/1812.11440. (Accessed 6 July C8MH00653A.
2023). [88] Z. Zhu, N. Anwer, Q. Huang, L. Mathieu, Machine learning in tolerancing for
[66] C.-H. Pham, A. Ducournau, R. Fablet, F. Rousseau, Brain MRI super-resolution additive manufacturing, CIRP Annals 67 (2018) 157–160, https://doi.org/
using deep 3D convolutional networks, in: 2017 IEEE 14th International 10.1016/j.cirp.2018.04.119.
Symposium on Biomedical Imaging (ISBI 2017), IEEE, Melbourne, Australia, [89] S. Chowdhury, S. Anand, Artificial neural network based geometric compensation
2017, pp. 197–200, https://doi.org/10.1109/ISBI.2017.7950500. for thermal deformation in additive manufacturing processes, in: Joint MSEC-
[67] J.E. Iglesias, B. Billot, Y. Balbastre, C. Magdamo, S.E. Arnold, S. Das, B.L. Edlow, NAMRC Symposia, American Society of Mechanical Engineers, vol. 3, 2016,
D.C. Alexander, P. Golland, B. Fischl, SynthSR, A public AI tool to turn V003T08A006, https://doi.org/10.1115/MSEC2016-8784. Blacksburg, Virginia,
heterogeneous clinical brain scans into high-resolution T1-weighted images for USA.
3D morphometry, Sci. Adv. 9 (2023), eadd3607, https://doi.org/10.1126/sciadv. [90] J. Francis, L. Bian, Deep learning for distortion prediction in laser-based additive
add3607. manufacturing using big data, Manufacturing Letters 20 (2019) 10–14, https://
[68] H.N. Monday, J. Li, G.U. Nneji, S. Nahar, M.A. Hossin, J. Jackson, C.J. Ejiyi, doi.org/10.1016/j.mfglet.2019.02.001.
COVID-19 diagnosis from chest X-ray images using a robust multi-resolution [91] M. Samie Tootooni, A. Dsouza, R. Donovan, P.K. Rao, Z. (James) Kong,
analysis siamese neural network with super-resolution convolutional neural P. Borgesen, Classifying the dimensional variation in additive manufactured parts
network, Diagnostics 12 (2022) 741, https://doi.org/10.3390/ from laser-scanned three-dimensional point cloud data using machine learning
diagnostics12030741. approaches, J. Manuf. Sci. Eng. 139 (2017), 091005, https://doi.org/10.1115/
[69] P. Tavousi, Z. Shahbazi, S. Shahbazmohamadi, AI-Based brain image 1.4036641.
segmentation using synthesized data, Microsc. Microanal. 26 (2020) 694–697, [92] J. Lee, S.J. Oh, S.H. An, W.-D. Kim, S.-H. Kim, Machine learning-based design
https://doi.org/10.1017/S143192762001555X. strategy for 3D printable bioink: elastic modulus and yield stress determine
[70] A. Crimi, S. Bakas, H. Kuijf, B.H. Menze, M. Reyes (Eds.), Brainlesion: Glioma, printability, Biofabrication 12 (2020), 035018, https://doi.org/10.1088/1758-
Multiple Sclerosis, Stroke and Traumatic Brain Injuries: Third International 5090/ab8707.
Workshop, BrainLes 2017: Held in Conjunction with MICCAI 2017: Quebec City, [93] S.V. Murphy, A. Skardal, A. Atala, Evaluation of hydrogels for bio-printing
QC, Canada, September 14, 2017: Revised Selected Papers, Springer, Cham, applications, J. Biomed. Mater. Res. 101A (2013) 272–284, https://doi.org/
2018. 10.1002/jbm.a.34326.
[71] L. Teng, H. Li, S. Karim, DMCNN: a deep multiscale convolutional neural network [94] J. Shin, Y. Lee, Z. Li, J. Hu, S.S. Park, K. Kim, Optimized 3D bioprinting
model for medical image segmentation, J. Healthcare Eng. 2019 (2019) 1–10, technology based on machine learning: a review of recent trends and advances,
https://doi.org/10.1155/2019/8597606. Micromachines 13 (2022) 363, https://doi.org/10.3390/mi13030363.
[72] J. Duan, M. Bernard, L. Downes, B. Willows, X. Feng, W.F. Mourad, W. St Clair, [95] K. Ruberu, M. Senadeera, S. Rana, S. Gupta, J. Chung, Z. Yue, S. Venkatesh,
Q. Chen, Evaluating the clinical acceptability of deep learning contours of G. Wallace, Coupling machine learning with 3D bioprinting to fast track
prostate and organs-at-risk in an automated prostate treatment planning process, optimisation of extrusion printing, Appl. Mater. Today 22 (2021), 100914,
Med. Phys. 49 (2022) 2570–2581, https://doi.org/10.1002/mp.15525. https://doi.org/10.1016/j.apmt.2020.100914.
[73] P.-J. Verhelst, A. Smolders, T. Beznik, J. Meewis, A. Vandemeulebroucke,
E. Shaheen, A. Van Gerven, H. Willems, C. Politis, R. Jacobs, Layered deep

13
L. Ma et al. Materials Today Bio 23 (2023) 100792

[96] A. Menon, B. Póczos, A.W. Feinberg, N.R. Washburn, Optimization of silicone 3D [111] R. Jafari-Marandi, M. Khanzadeh, W. Tian, B. Smith, L. Bian, From in-situ
printing with hierarchical machine learning, 3D Print. Addit. Manuf. 6 (2019) monitoring toward high-throughput process control: cost-driven decision-making
181–189, https://doi.org/10.1089/3dp.2018.0088. framework for laser-based additive manufacturing, J. Manuf. Syst. 51 (2019)
[97] J. Shi, J. Song, B. Song, W.F. Lu, Multi-objective optimization design through 29–41, https://doi.org/10.1016/j.jmsy.2019.02.005.
machine learning for drop-on-demand bioprinting, Engineering 5 (2019) [112] A. Caggiano, J. Zhang, V. Alfieri, F. Caiazzo, R. Gao, R. Teti, Machine learning-
586–593, https://doi.org/10.1016/j.eng.2018.12.009. based image processing for on-line defect recognition in additive manufacturing,
[98] I. Rojek, D. Mikołajewski, E. Dostatni, M. Macko, AI-optimized technological CIRP Annals 68 (2019) 451–454, https://doi.org/10.1016/j.cirp.2019.03.021.
aspects of the material used in 3D printing processes for selected medical [113] I.A. Okaro, S. Jayasinghe, C. Sutcliffe, K. Black, P. Paoletti, P.L. Green, Automatic
applications, Materials 13 (2020) 5437, https://doi.org/10.3390/ma13235437. fault detection for laser powder-bed fusion using semi-supervised machine
[99] B. Chen, J. Dong, M. Ruelas, X. Ye, J. He, R. Yao, Y. Fu, Y. Liu, J. Hu, T. Wu, learning, Addit. Manuf. 27 (2019) 42–53, https://doi.org/10.1016/j.
C. Zhou, Y. Li, L. Huang, Y.S. Zhang, J. Zhou, Artificial intelligence-assisted high- addma.2019.01.006.
throughput screening of printing conditions of hydrogel architectures for [114] L. Scime, J. Beuth, Using machine learning to identify in-situ melt pool signatures
accelerated diabetic wound healing, Adv. Funct. Mater. 32 (2022), 2201843, indicative of flaw formation in a laser powder bed fusion additive manufacturing
https://doi.org/10.1002/adfm.202201843. process, Addit. Manuf. 25 (2019) 151–165, https://doi.org/10.1016/j.
[100] S. Tian, R. Stevens, B. McInnes, N. Lewinski, Machine assisted experimentation of addma.2018.11.010.
extrusion-based bioprinting systems, Micromachines 12 (2021) 780, https://doi. [115] L. Scime, J. Beuth, Anomaly detection and classification in a laser powder bed
org/10.3390/mi12070780. additive manufacturing process using a trained computer vision algorithm, Addit.
[101] M. Rezapour Sarabi, M.M. Alseed, A.A. Karagoz, S. Tasoglu, Machine learning- Manuf. 19 (2018) 114–126, https://doi.org/10.1016/j.addma.2017.11.009.
enabled prediction of 3D-printed microneedle features, Biosensors 12 (2022) 491, [116] N. Razaviarab, S. Sharifi, Y.M. Banadaki, Smart additive manufacturing
https://doi.org/10.3390/bios12070491. empowered by a closed-loop machine learning algorithm, in: J. Kim (Ed.), Nano-,
[102] P. Kumar, P. Gupta, I. Singh, Parametric optimization of FDM using the ANN- Bio-, Info-Tech Sensors and 3D Systems III, SPIE, Denver, United States, 2019,
based whale optimization algorithm, AI EDAM (Artif. Intell. Eng. Des. Anal. p. 17, https://doi.org/10.1117/12.2513816.
Manuf.) 36 (2022) e27, https://doi.org/10.1017/S0890060422000142. [117] Y. Wang, J. Huang, Y. Wang, S. Feng, T. Peng, H. Yang, J. Zou, A CNN-Based
[103] D. Wu, C. Xu, Predictive modeling of droplet formation processes in inkjet-based Adaptive Surface Monitoring System for Fused Deposition Modeling, (n.d.).
bioprinting, J. Manuf. Sci. Eng. 140 (2018), 101007, https://doi.org/10.1115/ [118] W. Lee, J. Fritsch, A. Maqsood, S. Liu, T. Bourassa, R. Calara, W.S. Kim, Adaptive
1.4040619. 3D printing for in situ adjustment of mechanical properties, Adv. Intelligent Sys. 5
[104] I. Baturynska, O. Semeniuta, K. Martinsen, Optimization of process parameters for (2023), 2200229, https://doi.org/10.1002/aisy.202200229.
powder bed fusion additive manufacturing by combination of machine learning [119] D. Ye, J.Y. Hsi Fuh, Y. Zhang, G.S. Hong, K. Zhu, In situ monitoring of selective
and finite element method: a conceptual framework, Procedia CIRP 67 (2018) laser melting using plume and spatter signatures by deep belief networks, ISA
227–232, https://doi.org/10.1016/j.procir.2017.12.204. (Instrum. Soc. Am.) Trans. 81 (2018) 96–104, https://doi.org/10.1016/j.
[105] J. Go, A.J. Hart, Fast desktop-scale extrusion additive manufacturing, Addit. isatra.2018.07.021.
Manuf. 18 (2017) 276–284, https://doi.org/10.1016/j.addma.2017.10.016. [120] Y. Wang, Y. Lin, R.Y. Zhong, X. Xu, IoT-enabled cloud-based additive
[106] K. Paraskevoudis, P. Karayannis, E.P. Koumoulos, Real-time 3D printing remote manufacturing platform to support rapid product development, Int. J. Prod. Res.
defect detection (stringing) with computer vision and artificial intelligence, 57 (2019) 3975–3991, https://doi.org/10.1080/00207543.2018.1516905.
Processes 8 (2020) 1464, https://doi.org/10.3390/pr8111464. [121] S. Lee, M. Jeong, C.-S. Cho, J. Park, S. Kwon, Deep learning-based PC member
[107] T. Wang, T.-H. Kwok, C. Zhou, S. Vader, In-situ droplet inspection and closed-loop crack detection and quality inspection support technology for the precise
control system using machine learning for liquid metal jet printing, J. Manuf. construction of OSC projects, Appl. Sci. 12 (2022) 9810, https://doi.org/
Syst. 47 (2018) 83–92, https://doi.org/10.1016/j.jmsy.2018.04.003. 10.3390/app12199810.
[108] Z. Jin, Z. Zhang, G.X. Gu, Autonomous in-situ correction of fused deposition [122] P. Wang, H.-W. Tseng, T.-C. Chen, C.-H. Hsia, Deep convolutional neural network
modeling printers using computer vision and deep learning, Manufact. Letters 22 for coffee bean inspection, Sensor. Mater. 33 (2021) 2299, https://doi.org/
(2019) 11–15, https://doi.org/10.1016/j.mfglet.2019.09.005. 10.18494/SAM.2021.3277.
[109] J. Shi, B. Wu, B. Song, J. Song, S. Li, D. Trau, W.F. Lu, Learning-based cell [123] S. Sundaram, A. Zeid, Artificial intelligence-based smart quality inspection for
injection control for precise drop-on-demand cell printing, Ann. Biomed. Eng. 46 manufacturing, Micromachines 14 (2023) 570, https://doi.org/10.3390/
(2018) 1267–1279, https://doi.org/10.1007/s10439-018-2054-2. mi14030570.
[110] Z. Jin, Z. Zhang, X. Shao, G.X. Gu, Monitoring anomalies in 3D bioprinting with [124] B. Jung, H. You, S. Lee, Anomaly Candidate Extraction and Detection for
deep neural networks, ACS Biomater. Sci. Eng. (2021), 0c01761, https://doi.org/ automatic quality inspection of metal casting products using high-resolution
10.1021/acsbiomaterials.0c01761 acsbiomaterials. images, J. Manuf. Syst. 67 (2023) 229–241, https://doi.org/10.1016/j.
jmsy.2023.02.007.

14